Database Study

I have started playing some daily games and attempting to use the database in them.

I have reached a case, which I can't discuss specifically since the game is ongoing.  But basically, there's a situation where a particular move is scoring 50%.  But if you look at the opponents next move, he's got a lot of choices.  One of his options has a 100% score out of three master-level games, yet it hasn't been played since 2004.  Another of his options has an 83% score, also out of three master level games, yet it hasn't been played since 1950.  Instead, a different move is heavily preferred, including by well-known players, even though it scores much worse.  Now, it might be assumed that "oh, that's because the move has been refuted," but if so I can't find the game where it happened.

So--what gives?  Obviously it would help if I could show the line.  Or if I could turn on an engine.  But I can't do that, at least not right now.  I am just wondering, in general, do you run into this sort of thing very often?  And if so, what is going on?

Your logic is hogwash. You are basing these 100% and 83% anomalies on lines with 3 games. You need to learn the laws of statistics. To be statistically relevant, you need at least a sample size of 30. So your samples of 3 games are statistically irrelevant!

It must be nice to be so confident. All 30 really does for you is get you comfortably within a z-distribution instead of being stuck with the inferior t-distribution. Even that is an over-generalization. There's a whole statistical field dealing with small sample sizes. Even more important is that the results of a chess game given a certain position aren't random anyway. Other than that, your thesis that my logic is hogwash is well demonstrated.

For the case in question, if we have a position that has occurred 28 times at master level, with the most preferred move played 15 times with 5 wins, 7 losses, and 3 draws, and a less preferred move played 3 times with 3 wins, that would seem to be worthy of further inquiry.

I only asked whether such circumstance is seen frequently. If it's not of interest to you, certainly you don't owe any answer. There's no call to say my logic is hogwash. And then, of course, it isn't anyway.

Edit: removed bold from text...not sure how it got there.

You can spew all you want. I actually took multiple college level classes on statistics, and in chess, it's just as relevant.

The move sequence 1.e4 e6 2.d4 d5 has occurred thousands upon thousands of times. It scores like any other sound opening for Black - somewhere in the mid-to-upper 40s. Typically 46% is the average score for Black. The move sequence 1.e4 Nh6 2.d4 f5 is far less often played, and let's hypothetically say that it has been played 17 times with 6 wins for White, 9 wins for Black, and 2 draws.

The numbers are statistically irrelevant because your sample size is too low and your Variance/Standard Deviation is way to high with that few games. But why do so many still play the French, Sicilian, Caro-Kann, and 1...e5 against 1.e4 rather than some bogus line that happens to be a plus score for Black based on so few games? It's because statistics are more than percentages. It's variance as well Your 3 games where Black went 3-0 has so high of a variance that it becomes statistically irrelevant.

Wow...multiple college level courses. And yet, you aren't aware that 30 as a minimum sample size is just a basic rule of thumb taught to freshmen. One that applies--and then only minimally--to z-scores on normal distributions. It is a lie told to children. You have no reason to assume it applies to this situation. There are many valid studies conducted with far fewer measures, even single digits. And, of course, the whole discussion loses meaning in situations which don't involve randomness anyway.

More concretely, do some people really play the French, Sicilian, Caro, etc., because of the reason you state which was "it's because statistics are more than percentages." Of course not. They play it because they hope they know it better.

And as to whether "it's variance" and "statistically irrelevant" that there are 3-0 games. It does not (at least not clearly) have anything to do with variance (from some implied mean). And it's not statistically irrelevant. Because it doesn't matter what the z-score is. Because that concept doesn't apply here. It is at minimum a low-confidence result worthy of some consideration.

Moreover, it's highly unlikely that any GM who plays the line is unaware that it has a recent win by a 2600+ player and two earlier wins as well (and no losses or draws). But they don't play it. That's got to be more important to the case at hand than statistics.

But, I guess I'll have to wait for this game to be over so I can put an engine on it. Because we aren't getting anywhere here.